Raise the grid to Excel's 1,048,576 rows by 16,384 columns

Columns now run to XFD, with three-letter names in references, and
formulas take whole columns (A:C, $A:$A) and rows (2:5), which stay
whole when copied or when lines are inserted and deleted; ranges print
that way, A1:A1048576 as A:A.

For the engine to cost what the data costs at that size:

- A second occupancy index holds the cells with contents, so the used
  range, data-edge jumps and the next filled column of a row are found
  by index rather than by walking a million rows. A filter over whole
  columns tests the rows with data and one blank row for the rest, and
  steps over hidden blank rows at once.
- Formatting whole columns or rows keeps the format on the line, as
  Sheets does, rather than materializing a cell per address: a cell
  shows its own format, else its row's, else its column's, part by part;
  one the fallback can't express keeps its own whole. Line formats undo,
  move with inserted lines and are saved in an additive "lines" field.
- Paste clears what the clip's blanks land on and places each clip cell
  per tile; fills of blank lanes only clear; clipboard text stops at the
  last value; pivots and JEV read only their data rows.
- Pastes and fills are capped by a max-cells budget (two million by
  default) instead of a fixed 1 M.

commit

1c459ce04e8eeda23b0cb4e64fee22ebf4ca5219

parent

c114f3207a388da596438aa314fe99d443c757df

author

Dami <dami.etoile@gmail.com>